https://github.com/cukas/claude-workflow-skills
Guarded development workflows for Claude Code — investigation before fixes, orchestration before implementation, verification before completion
https://github.com/cukas/claude-workflow-skills
claude-code claude-code-plugin claude-code-skills code-quality developer-tools tdd
Last synced: 2 months ago
JSON representation
Guarded development workflows for Claude Code — investigation before fixes, orchestration before implementation, verification before completion
- Host: GitHub
- URL: https://github.com/cukas/claude-workflow-skills
- Owner: cukas
- License: mit
- Created: 2026-03-09T08:25:55.000Z (4 months ago)
- Default Branch: main
- Last Pushed: 2026-03-09T14:01:40.000Z (4 months ago)
- Last Synced: 2026-03-09T15:51:20.496Z (4 months ago)
- Topics: claude-code, claude-code-plugin, claude-code-skills, code-quality, developer-tools, tdd
- Homepage:
- Size: 11.7 KB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Claude Workflow Skills
Guarded development workflows for Claude Code. Investigation before fixes. Orchestration before implementation. Verification before completion.
> **Pure discipline, zero infrastructure.** No databases, no MCP servers, no build tools. Just markdown skills that teach Claude to work methodically — read before edit, test before ship, verify before done. Works standalone. Layers with [patrol](https://github.com/cukas/patrol) and [remembrall](https://github.com/cukas/remembrall) if installed.
## Skills
### `/build-guard`
Orchestrated multi-file implementation with regression checkpoints.
1. **Plan** — enters plan mode (read-only), explores the codebase, writes a numbered plan with risks
2. **Implement** — dispatches subagents for larger changes (one per file or logical chunk), type-checks after every file, runs tests every 3 files
3. **Verify** — auto-invokes `/review-gate` before declaring done
Auto-scales: small changes (1-2 files) run inline, larger changes use subagents to keep context clean. For big plans that consume context, hands off via remembrall and resumes with full budget.
```
/build-guard → plan mode → approval → subagent dispatch → checkpoints → /review-gate
```
### `/trace-fix`
Investigation-first bug fixing with TDD loop.
1. **Trace** — reads every file in the call chain, documents the root cause before touching code
2. **Fix** — writes a failing test, implements the minimal fix, runs full test suite
3. **Verify** — auto-invokes `/review-gate` when green
Escapes after 5 failed iterations instead of spinning forever. Falls back gracefully when no test framework or type checker exists.
```
/trace-fix → trace root cause → failing test → minimal fix → verify → /review-gate
```
### `/review-gate`
Structured post-implementation review with pass/fail output.
Traces affected code paths, scans for bug patterns (null safety, async errors, security, stale data), runs build/type/test verification, checks architectural consistency. Outputs a structured report:
```
| Check | Status | Notes |
|------------------|--------|----------------------------|
| Code path trace | PASS | 3 callers, 2 dependents |
| Bug pattern scan | PASS | No issues found |
| Build | PASS | npm run build |
| Type check | PASS | npx tsc --noEmit |
| Tests | PASS | 47 passed, 0 failed |
| Architecture | PASS | Consistent with src/utils |
```
Both `/build-guard` and `/trace-fix` auto-invoke this — nothing ships without passing the gate.
## Install
```bash
claude plugin marketplace add cukas/claude-workflow-skills
claude plugin install workflow-skills@cukas
```
## Works With
These skills are **self-contained** — they enforce discipline on their own. But they integrate smoothly with:
- **[patrol](https://github.com/cukas/patrol)** — If installed, build-guard and trace-fix respect patrol's investigation warnings and escalation. If not, the skills enforce the same read-before-edit discipline themselves.
- **[remembrall](https://github.com/cukas/remembrall)** — If installed, build-guard hands off context-heavy sessions automatically for clean context on resume. If not, plans are saved to `docs/plans/` as files.
## License
MIT